recipe.yaml

quant_stage:
  quant_modifiers:
    GPTQModifier:
      config_groups:
        group_0:
          targets: [Linear]
          weights: {actorder: weight, group_size: 128, num_bits: 4, observer: mse, strategy: group,
            symmetric: true, type: int}
      dampening_frac: 0.01
      ignore: [lm_head]